Metric Spaces

ii Dr. V. V.
Acharya
Topology of Metric Spaces

Textbook written according to the latest
syllabus of University of Pune
Dr. V. V. Acharya
Head, Department of Mathematics,
Fergusson College, Pune-411 004.
January 14, 2015

iv Dr. V. V. Acharya
Preface
Contents
These are the lecture notes given by me at Fergusson College
n n based on the book by Prof. S. Kumaresan. These notes are useful
1 R, R and C, C 1
for students who want to study metric spaces. An attempt has
1.1 R. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
been made to make chapters small so that the student should
1.2 Rn . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
understand that concept. There is an emphasis on examples.
1.3 Complex Numbers . . . . . . . . . . . . . . . . . . . . . . . 3
1.4 Cn . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
1.5 Some Important Inequalities . . . . . . . . . . . . . . . . . . 6 The aim of the course is to prove the generalization of
the following: theorem: If f : [a, b] R is a continuous
2 Metric Spaces 11 function then f ([a, b]) is a closed and bounded interval.
Suggestions for improvement of notes are welcome. The sugges-
3 Open and closed sets in Metric spaces 15 tions may be sent at vvacharya@gmail.com.
3.1 Open Balls . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
3.2 Open sets . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
3.3 Closed sets . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
3.4 Sequences in a metric space . . . . . . . . . . . . . . . . . . 20
4 Limits in Metric spaces 25 Dr. V. V. Acharya

4.1 Continuous functions . . . . . . . . . . . . . . . . . . . . . . 26
4.2 Equivalent Definitions of Continuity . . . . . . . . . . . . . 27
5 Compact Metric Spaces 33

5.1 Compact Set . . . . . . . . . . . . . . . . . . . . . . . . . . 33
5.2 Bounded and Totally Bounded sets . . . . . . . . . . . . . . 33
5.3 Complete Metric spaces . . . . . . . . . . . . . . . . . . . . 37
5.4 Equivalent conditions for Compact metric spaces . . . . . . 39
5.5 Continuous functions on compact metric space . . . . . . . 42
5.6 Continuity of the inverse function . . . . . . . . . . . . . . . 43
5.7 Uniform Continuity . . . . . . . . . . . . . . . . . . . . . . . 44
6 Connected Spaces 47
iii
2 Dr. V. V. Acharya
2. kkxk = |k| kxk

3. kx + yk kxk + kyk .
Chapter 1 Proof. We observe that 1 and 2 are obvious. To prove 3, we need Cauchy-
Schwartz Inequality.
R, Rn and C, Cn Exercise 1.1 Let a1 , . . . , an ; b1 , . . . , bn be real numbers. Find the discrim-

inant of the quadratic equation given by
1.1 R (a1 x b1 )2 + (a2 x b2 )2 + + (an x bn )2 = 0.
Let R denote
( the set of real numbers. We define, the absolute value of x by Show that the discriminant of this equation is always non-negative.
x if x 0
|x| = We have seen the following properties of absolute Theorem 1.4 (Cauchy-Schwartz Inequality) Let a1 , . . . , an and
x if x < 0.
b1 , . . . , bn be any real numbers. Then
value function:
(a21 + . . . + a2n )(b21 + . . . + b2n ) (a1 b1 + . . . + an bn )2 (1.1)
Theorem 1.1 1. |x| 0. Further, |x| = 0 if and only if x = 0.
2. |kx| = |k||x| and equality occurs if and only if either
3. |x + y| |x| + |y|. 1. all the ai s are zero or all the bi s are zero or
2. the ai s are proportional to the bi s i.e. there exists k 6= 0 such that
If x, y R then we define distance between x and y by d(x, y) = |x y|.
ai = kbi for all i.
Theorem 1.2
Proof. Follows from exercise 1.1.
1. d(x, y) > 0 if x 6= y and d(x, y) = 0 if x = y. coincide. That is,
d(x, y) 0 for all x, y R and d(x, y) = 0 if and only if x = y. Proof of (3). Note that
n n n n
2. d(x, y) = d(y, x), i.e. distance is symmetric.
kx + yk2
X X X X
= (xi + yi )2 = x2i + yi2 + 2 xi yi
3. If x, y, z are any three points, then d(x, y) d(x, z) + d(z, y). i=1 i=1 i=1 i=1
Xn n
X n
X
Proof. This follows using the properties of the absolute value. x2i + yi2 + 2 |xi yi |
i=1 i=1 i=1
v v
1.2 Rn n
X n
X
u n
X uX
u n
x2i + y 2 + 2t
u
i x2 t y2 i i
n
Note that R = {(x1 , x2 , . . . , xn )|xi R} . Let x = (x1 , x2 , . . . , xn ), y = i=1 i=1 i=1 i=1
(y1 , y2 , . . . , yn ) Rn . We define, x + y = (x1 + y1 , . . . , xn + yn ) and v
u n
v
u n
2
p = (kx1 , . . . , kxn ). Further, the length of x, kxk , is defined by kxk =
kx

uX uX
x2i + t yi2 = (kxk + kyk)2 .
x21 + + x2n . Length of x is also called as norm of x. The norm of x has
t
i=1 i=1
the following properties:
Theorem 1.3 1. kxk 0. Further, kxk = 0 if and only if x = 0. Hence, kx + yk kxk + kyk .
1
Modern Analysis 3 4 Dr. V. V. Acharya
If points A and B on the real axis have co-ordinates a and b respectively, i = (0, 1) is the imaginary unit and i2 = 1. Also x is called the real part
then the distance between them is d(A, B) = |a b|. and y the imaginary part of the complex number z = x + iy and we write
x = Re(z), y = Im(z). Note that since x, y are real numbers, we have
We define the distance between two points in a plane and space as usual:
the distance between the points A(a, b) and P (x, y) in R2 is denoted as x2 0 and y 2 0. (1.5)
d(A, P ) and is defined thus:
p Further, z = x iy is called the conjugate of z and
d(A, P ) = + (x a)2 + (y b)2 . (1.2) p
|z| = + x2 + y 2 (1.6)
Similarly, the distance between the points A(a, b, c) and P (x, y, z) in R3 is
defined as follows: is called the modulus of z, so that z z = |z|2 . By (1.5) we see thatp |z| =
p 0 z =p 0 and also x2 x2 + y 2 and y 2 x2 + y 2 . Hence, |x| x2 + y 2
d(A, P ) = + (x a)2 + (y b)2 + (z c)2 . (1.3) and |y| x2 + y 2 . Thus,
More generally, the distance between the points A(a1 , . . . , an ) and |Re(z)| |z| and |Im(z)| |z| (1.7)
P (x1 , . . . , xn ) in Rn is defined by
p Given any complex numbers z, w we have the triangle inequality
d(A, P ) = + (x1 a1 )2 + + (xn an )2 . (1.4)
|z w| |z| + |w| (1.8)
Note that the distance in Rn as defined above is in fact given by d(A, P ) =
kx ak . Further, it has the following basic properties: which also implies the inequality
Theorem 1.5 1. d(A, P ) > 0 if A, P are distinct points and d(A, P ) = ||z| |w|| |z w| (1.9)
0 if A, P coincide. That is, d(A, P ) 0 always and d(A, P ) = 0 if
and only if A, P coincide. Since, |i| = 1 it follows from (1.8) that
2. d(A, P ) = d(P, A), i.e. distance is symmetric. |x + iy| |x| + |iy| = |x| + |i||y| = |x| + |y|. Hence,
3. If A, B, P are any three points, then d(A, P ) d(A, B) + d(B, P ). |z| |Re(z)| + |Im(z)|. (1.10)
Proof. Note that d(A, P ) = kx ak . Hence, using Theorem 1.3 we get These simple results are of great importance.
the result. Consider rectangular coordinate axes OX, OY in a plane. Then we can
Or If A(a, b), P (x, y) R2 , then property (1.2) says that AP = 0 if and interpret the points in the plane in two ways, depending on our point of
only if AP 2 = 0 if and only if (x a)2 + (y b)2 = 0 if and only if x = a view as follows:
and y = b. Also, if A, P are distinct, then either x 6= a or y 6= b i.e. either To every point in the plane corresponds a unique ordered pair (x, y) of
(x a)2 > 0 or (y b)2 > 0 so that AP > 0. In the similar way, we can co-ordinates of P and also to P there corresponds a unique complex number
prove the general case. Prove it! z = x + iy. In both cases the correspondence is one-to-one. Therefore in
the first case we identify the plane with the Cartesian product R2 = R R
1.3 Complex Numbers and call it the real plane. In the second case we identify the plane with the
set C and call it the complex plane (or Argand plane or Gaussian plane).
Let C denote the set complex numbers. Throughout this section x,y,a,b, Thus the point P (x, y) in the real plane corresponds to the point P (z) in
c,d, will denote real numbers. Recall that complex number z is an ordered the complex plane. Also, in the complex plane, the X and Y axes are called
pair of real numbers, z = (x, y), and is also written as z = x + iy. Here the real axis and the imaginary axis respectively.
Further, if points A, B represent complex numbers z = x + iy and 2. d(A, P ) = d(P, A), i.e. distance is symmetric.
w = a + ib in the complex plane C, we define the distance d(A, B) between
them to be |z w|, i.e. 3. If A, B, P are any three points, then d(A, P ) d(A, B) + d(B, P ).
Proof. Note that d(A, P ) = kx ak . Hence, using Theorem 1.6 we get
the result.
p
|z w| = d(A, B) = + (x a)2 + (y b)2 . (1.11)
Thus the distance d(A, B) in C is the same as between the corresponding

points (x, y) and (a, b) in R2 . Hence, we get the following properties of the
1.5 Some Important Inequalities
distance function: The Three Means: If a, b are positive real numbers, we define their arith-
1. d(A, B) 0 for all A, B C. metic mean (A.M.), geometric mean (G.M.) and harmonic mean (H.M.) as
follows
2. d(A, B) = 0 if and only if A = B. 1

1 1
a+b a + b
3. d(A, B) = d(B, A) for all A, B C A.M. = A = , G.M. = G = + ab, H.M. = H = .
2 2
4. d(A, B) d(A, C) + d(C, B) for all A, B, C C If a1 , . . . , an are n positive real numbers, we define their arithmetic mean
An , geometric mean Gn and harmonic mean Hn as follows:
1.4 Cn
a1 + . . . + an n
An = , Gn = (a1 a2 . . . an )1/n , Hn = 1 1 .
n
Note that C = {(z1 , z2 , . . . , zn )|zi C} . Let z = (z1 , z2 , . . . , zn ), w = n a1 + ...+ an
(w1 , w2 , . . . , wn ) Cn . We define, z + w = (z1 + w1 , . . . , zn + wn ) and
kz = (kz1 , . . . , kzn ). Further, the length of z, kzk , is defined by kzk = Remark 1.1 Note that the H.M. is the reciprocal of the A.M. of the re-
2ab
z1 z1 + + zn zn . Length of z is also called as norm of z. The norm of z ciprocals of the given numbers. Further, H.M. of a and b is . We also
has the following properties: a+b
note that the A.M. can be defined for any real numbers (not necessarily
Theorem 1.6 1. kzk 0. Further, kxk = 0 if and only if z = 0. positive) but to define G.M. and H.M. we require necessarily positive real
numbers.
2. kkzk = |k| kzk
Theorem 1.8 If a, b are positive real numbers, then A G H. Also,
3. kz + wk kzk + kwk . equality occurs if and only if a = b.

The distance between the points A(a1 , . . . , an ) and P (z1 , . . . , zn ) in Cn is Proof: First let a 6= b. Then a and b are unequal and so ( a b)2 > 0.
a+b
defined by Hence the identities A G = ab = 12 ( a b)2 , G H =
2
p
d(A, P ) = + |z1 a1 |2 + + |zn an |2 . (1.12) 2ab ab( a b)2
ab = respectively show that A > G and G > H.
a+b a+b
Note that the distance in Cn as defined above is in fact given by d(A, P ) = If a = b, it is clear that A = G = H = a.
kz ak . Further, it has the following basic properties:
Theorem 1.9 If a1 , . . . , an are positive real numbers, then An Gn
Theorem 1.7 1. d(A, P ) > 0 if A, P are distinct points and d(A, P ) = Hn , and equality occurs if and only if a1 = a2 = . . . = an .
0 if A, P coincide. That is, d(A, P ) 0 for all A, P Cn and Proof: First we prove by induction on n that An Gn with equality if and
d(A, P ) = 0 if and only if A, P coincide. only if all the n numbers are equal. The result is true for n = 2. Assume
the result for n = m 1. Suppose 0 < a1 a2 . . . am . We note that f (x0 ) is the minimum of f on (0, ). Now, 1
the result is true if a1 = am . Suppose a1 < am . Then clearly 1 p
xp0 yq y p1 yq 1
ma1 < a1 + a2 + . . . + am < mam , f (x0 ) = + x0 y = + y p1 y (1.15)

p q p q
yq yq

1 1 1
so that a1 < Am < am . Hence = + y p1 y = y q + y q = 0. (1.16)
p q p q
Am (a1 + am Am ) a1 am = (a1 Am )(Am am ) > 0 and so
Thus, f (x) 0 = f (x0 ) for all x (0, ), which is the required inequality.
a1 am 1 p
Further, equality occurs if and only if x = y p1 i.e. xp = y p1 = y q .
a1 + am Am > (1.13)
Am
Theorem 1.11 (Holders Inequality) Let K denote Pn R or C. For x =
Now since the A.M. of the m 1 numbers a2 , . . . , am1 , a1 + am Am is (x1 , . . . , xn ) Kn and 1 p < , let kxkp = ( i=1 |xi |p )1/p , and let
Am , we have by induction hypothesis, kxk = max{|xi | : i = 1 . . . n}. For p > 1, let q be defined in such away
1 1
a1 am that + = 1 holds and for p = 1 let q = . Then
m1
Am a2 . . . am1 (a1 + am Am ) > a2 . . . am1 [by (1.13)] p q
Am X
|ai ||bi | kakp kbkq , for all a, b Kn . (1.17)
Hence Am m > a1 a2 . . . am i.e. Am > Gm .
i
Thus the result is true for n = m and the induction is complete. We Equality holds if and only if C1 |ak |p = C2 |bk |q for 1 k n for some
also observe that when the ai s are unequal, the arithmetic mean is strictly non-zero constants C1 and C2 .
greater than the geometric mean. Hence, if the arithmetic mean and the |ai | |bi |
Proof. Let x = ,y = . Then using (1.14), we get
geometric mean are equal then all the ai s must be equal. Finally, applying kakp kbkq
this result to the n positive numbers 1/a1 , . . . , 1/an we see that Gn Hn ,
with equality if and only if a1 = . . . = an . |ai | |bi | 1 |ai |p 1 |bi |q
p + (1.18)
kakp kbkp p kakp q kbkqq
Theorem 1.10 (Youngs Inequality) Let x, y be non-negative real num- Summing (1.18) for i = 1 to i = n, we get
1 1
bers. Let p > 1 and q be defined in such a way that + = 1 holds. Then n n
! n
!
p q X |ai | |bi | 1 X |ai |p 1 X |bi |q
p + q .
kakp kbkq p kakp q kbkq
xp yq i=1 i=1 i=1
xy + . (1.14) Simplifying, we get
p q n
X |ai ||bi | 1 1
p q + ,
Equality holds if and only if x = y . kak p kbk p p q
i=1
Proof. Fix y > 0. Define the function
whence we get (1.17). It is easy to see that the equality holds if and only
xp
y q if equality holds in Youngs inequality i.e. if and only if
f (x) = + xy for x > 0.
p q |ai |p |bi |q
p = q for 1 k n.
kakp kbkq
The derivative of f (x) is f (x) = xp1 y. Equating f (x) to 0, we get the
1 11 1 1 p1 p 1
critical point as x0 = y p1 . Now,f (x0 ) = (p 1)xp2 > 0, as p > 1. Thus, p
+
q
=1 q
= p
q= p1
=1+ p1
.
Thus, equality holds if and only if C1 |ak |p = C2 |bk |q for 1 k n for some
non-zero constants C1 and C2 . n
X
|ai ||ai + bi |p1 kakp |ai + bi |p1 q (using 1.17)

Now,
Remark 1.2 If we take p = 2 then q = 2. In this case, we get i=1
n
!1/q
p1 q
X X
|ai ||bi | kak2 kbk2 , for all a, b Kn . = kakp |ai + bi |
i i=1
n
!1/q
n
X
Thus, we get Cauchy Schwartz inequlity in K . = kakp |ai + bi |(p1)q (1.21)
i=1
!1/q
With the notation as in Holders Inequality, we have the following: n
X
= kakp |ai + bi |p
Theorem 1.12 (Minkowskis Inequality) Let 1 p . Then i=1
= kakp ka + bkp/q
p (1.22)
n
ka + bkp kakp + kbkp for a, b K . (1.19) n
p/q
X
Similarly, |bi ||ai + bi |p1 kbkp ka + bkp (1.23)
Equality holds if and only if there exist some non-zero constants C1 and C2 i=1
such that C1 a = C2 b.
Proof. We will consider three cases. Using (1.20),(1.22) and (1.23), we get
Case 1. p = . For a, b Kn , we have ka + bk = max{|ai + bi | : p p/q p/q
i = 1 . . . n}. Now, using triangle inequality, we get |ai + bi | |ai | + |bi | ka + bkp kakp ka + bkp + kbkp ka + bkp . (1.24)
kak + kbk . Thus,
Dividing both sides of (1.24) by ka + bkp/q
p , we get (1.19.)
2
ka + bk kak + kbk . Remark 1.3 1. Recall that, in an inner product space, we used Cauchy
Pn Scwartz inequality to establish triangle inequality for norm.
Case 2. p = 1. For a, b Kn , we have ka + bk1 = i=1 |ai + bi |. Now,
using triangle inequality, we get |ai + bi | |ai | + |bi |. Thus, 2. Minkowskis Inequality for p = 2 gives us triangle inequalty in the
inner product space Kn .
ka + bk1 kak1 + kbk1 .
Case 3. 1 < p < . It is easy to see that the result holds if either a or b
equals 0. Hence, we may assume that neither a nor b equals zero. Now,
n n
p
X X
ka + bkp = |ai + bi |p = |ai + bi ||ai + bi |p1
i=1 i=1
n
X n
X
|ai ||ai + bi |p1 + |bi ||ai + bi |p1 (1.20) 1
i=1 i=1
2 We use the following facts: 1. p = q(p1) and q
= p1
p
and 2. p pq = p(1 q1 ) = 1.
12 Dr. V. V. Acharya
7. Let M = C. Define d(z, w) = |z w|. It is easy to see that (C, d)

is a metric space. Further, if M = Cn then we can generalize the
examples 4, 5 and 6.
Chapter 2
8. Let V be a vector space with a norm k k. (V is called a normed linear
space). Then, V is metric space if we define d(v1 , v2 ) = kv1 v2 k for
Metric Spaces all v1 , v2 V.
Thus, in particular, every inner product space is a metric space as
Definition 2.1 Let X be any non-empty set. A function d : X X R every inner product space is a normed linear space. Examples 1, 3
is said to be a metric or a distance function defined on X if d satisfies the and 6 are thus special cases of this as R, Rn , C are inner product
following the following properties: spaces. Note that we will get now many more interesting examples of
d(x, x) = 0 (x X) metric space using a normed linear space or an inner product space.
d(x, y) > 0 (x, y X, x 6= y) Exercise 2.1 Are the following objects metric spaces, or not?
d(x, y) = d(y, x) (x, y X) 1. The Euclidean plane with points x = (xl , x2 ), but with distance de-
fined by d(x, y) = |x1 y1 |.
d(x, y) d(x, z) + d(y, z) (x, y, z X) (Triangle inequality)
If d is a metric for X, then the ordered pair (X, d) is called a metric space. 2. The set of cities in the United States that have airports, with
d(A, B) = scheduled air travel time from A to B.
Example 2.1 1. Let M = R. Define d(x, y) = |x y|. It is easy to see
that (R, d) is a metric space. 3. The positive real numbers with d(x, y) = x/y.
4. As in (3) but with d(x, y) = | log(x/y)|.

(
6 y,
1 if x =
2. Let M = R. Define d(x, y) =
0 if x = y. 5. The set of all positive numbers represented by terminating decimals,
Then (R, d) is a metric space. with d(x, y) = |x y| rounded to 10 decimal places.
(
1 if x 6= y, Assignment 1
3. Let M be any non-empty set. Define d(x, y) =
0 if x = y.
It is easy to see that (M, d) is a metric space. This is called as a 1. Define metric space.
discrete metric space.
2. Give two examples of metric spaces.
n
4. Let M = R . Define d2 (x, y) = kx yk . We have already proved (in
the last chapter) that (Rn , d) is a metric space. 3. Suppose d is a non-negative real valued-function defined for arbitrary
x and y in a set X and satisfies the conditions.
5. Let M = Rn . Define d (x, y) = max{|xi yi |}. Then (Rn , d ) is a (i) d(x, y) = 0 if and only if x = y.
metric space. (ii) d(x, y) d(z, x) + d(z, y) for arbitrary x, y, z in X. Show that d
n is a metric on X.
X
6. Let M = Rn . Define d1 (x, y) = |xi yi |. Then (Rn , d1 ) is a metric
4. Show that the set C of all complex numbers with d(z1 , z2 ) = |z1 z2 |,
i=1
space. z1 , z2 C is a metric space.
11
5. Let C be the set of all complex numbers. For x, y C define 16. Let (X, d) be a metric space. Show that the function d : X X R
2|x y| defined by d (x, y) = inf{1, d(x, y)}, x, y X is also a metric.
d(x, y) = p p . Show that (C, d) is a metric
{1 + |x|2 } {1 + |y|2 }
space. 17. For a metric space (X, d), prove that |d(x, y) d(x, z)| d(y, z).
Interpret this relation with usual metric in the plane.
6. (X, d) is a metric space. Define d(x, y) = C d(x, y) for all x, y X
and 0 < C < 1. Show that d is also a metric on X. 18. If M denotes the set of all n n matrices over the reals, examine
whether the following real valued function d defined on M M is a
d(x, y)
7. If d(x, y) is a metric on X, show that (x, y) = is also a metric on M : d(A, B) = | det A det B|, A, B M.
1 + d(x, y)
metric on X. 19. Define : R R R by (x, y) = |x+ y|. Is a metric for R? Justify!
2 2
8. Let the mapping d : R R R be defined by 20. Let (X, d1 ), (Y, d2 ) be metric spaces and M = X Y. Define
p
d(x, y) = (x1 x2 )2 + (y1 y2 )2 ((x1 , y1 ), (x2 , y2 )) = max{d1 (x1 , x2 ), d2 (y1 , y2 )}
2 2 2 ((x1 , y1 ), (x2 , y2 )) = d1 (x1 , x2 ) + d2 (y1 , y2 )
where x = (x1 , y1 ) R , y = (x2 , y2 ) R . Prove that (R , d) is a
metric space.
q
and d((x1 , y1 ), (x2 , y2 )) = d21 (x1 , x2 ) + d22 (y1 , y2 )
9. Let the mapping d : R2 R2 R be defined by d(x, y) = |x1
x2 | + |y1 y2 | where x = (x1 , y1 ) R2 , y = (x2 , y2 ) R2 . Prove that Show that , as well as d are metric on M.
(R2 , d) is a metric space. We call as a product metric.
10. Let the mapping d : R2 R2 R be defined by d(x, y) = max{|x1 Solutions

x2 |, |y1 y2 |} where x = (x1 , y1 ) R2 , y = (x2 , y2 ) R2 . Prove that
(R2 , d) is a metric space. 19. Note that (1, 1) = 0 but 1 6= 1. Hence, is not a metric for R.
11. Let d1 and d2 be two metrics for a non-empty set X. Then d defined
by d(x, y) = d1 (x, y) + d2 (x, y) is also a metric for X.
12. Let (X, d) be a metric space. Show that for arbitrary x, y, z, t in X,
|d(x, y) d(z, t)| d(x, z) + d(y, t).
13. Show that the set Rn of all n-tuples of real numbers with d(x, y) =
n
X
[ (x1 y1 )2 ]1/2 x, y Rn is a metric space.
1
14. Show that the set of continuous real valued functions defined on [0, 1]
with d(f, g) = sup |f (x) g(x)| is a metric space.
0x1
15. Let X be the set of all continuous

Z 1 real valued functions defined on
[0, 1]. For f X, define kf k = |f (x)| dx. Show that the mapping
0
d : X X R, defined by d(f, g) = kf gk, is a metric for X.
Exercise 3.2 Determine B((0, 0); 1) in the following metric spaces:

1. (R2 , d2 ), the usual Euclidean metric.
Chapter 3 2. (R2 , d ), the sup or the max norm.
3. (R2 , d1 ), the L1 norm.
Open and closed sets in Metric spaces
Exercise 3.3 In the metric space, (R2 , d ), determine B[(0, 0); 1].
3.1 Open Balls Exercise 3.4

In themetric Rd , that
space is R with discrete metric, de-
3 1 3 3 3 3
termine B ; ,B ; ,B ;1 ,B ;2 .
Definition 3.1 Let (M, d) be a metric space. If a M and r > 0, then 4 2 4 2 4 4
B(a; r) is defined as the set of all points in M whose distance from a is less Exercise 3.5 Find all the open balls in metric space (R, d).
than r. That is,
B(a; r) = {x M |d(x, a) < r}.
3.2 Open sets
We call B(a; r) the open ball of radius r about a.
B[a; r] is defined as the set of all points in M whose distance from a is less Definition 3.2 Let S be a subset of the metric space M. A point a M
than or equal to r. That is, is called an interior point of S if there exists r > 0 such that B(a, r) S.
The set of all the interior points of S is denoted by S 0 and is called interior
B[a; r] = {x M |d(x, a) r}. of S.
We call B[a; r] the closed ball of radius r about a. Definition 3.3 Let G be a subset of the metric space M. G is said to be
an open subset of M if for every x G there exists > 0 such that B(x; )
lies entirely in G.
Remark 3.1 Let G be a subset of the metric space M. G is said to be an
b+ b
open subset of M if every point of G is an interior point.

b b
b b b
Theorem 3.1 Let (M, d) be a metric space and a M. If r > 0 then
(a, b) (a, b) B(a; r) is open in (M, d).
b b
b b b b b b
O a O a a a+
Figure 3.1 Figure 3.2
2
Figure 3.1 shows the open ball B((a, b); r) in (R , d2 ) while Figure 3.2 shows
the open ball B((a, b); d ) in (R2 , d ).
Exercise 3.1Consider
the metric
space
[0, 1]
with usual
metric.
1 1 1 1 3 1
Determine B ; 1 , B 0; ,B ; ,B ; B(0; 2) in C
2 2 4 2 4 2
15
Proof. Let x B(a; r) and s = d(a, x). Suppose t is a positive real number Proof. If x G, then there exist an open interval B containing x such
less that r s. Consider B(x; t). If y B(x; t) then that B G. 1 Let
d(a, y) d(a, x) + d(x, y) s + t < r. S = {I G|I is an open interval containing x}.

S
Hence, y B(a; r). Hence, B(a; r) is open in (M, d). Since, B S, S is non-empty. Let Ix = IS I. Note that x Ix G.
Further Ix is an open set as Ix is the arbitrary union of open sets as every
Example 3.1 In a discrete metric space (M, d), singleton sets are open open interval is an open set. To show that Ix is an open interval it suffices
set. For, if x M then B(x; 1/2) = {x}. to show that if , Ix and < < then Ix . If = x then we are
done. Otherwise, there are two cases:
Theorem 3.2 In any metric space (M, d) the sets M and are both open. Case 1. < x. Since, Ix there exists an open interval I S such that
Proof. If x M then, by definition of B(x; r), every open ball is contained I. Hence, I as < < x and hence in Ix .
in M. Hence, M is open. The empty set is open simply because there is Case 2. > x. Since, Ix there exists an open interval I S such that
no x and hence for every x there exists r > 0 such that B(x; r) . I. Hence, I and x < < hence in Ix . Hence, Ix is an interval
containing x. Since, Ix is an open set, from the definition of Ix it follows
[ 3.3 If F be any family of open subsets of a metric space M,
Theorem that Ix is the largest open interval containing x.2 Hence,
then G is also an open subset of M. [
GF
[ G= Ix .
Proof. Let H = G. If H = then we are done. Assume H 6= and xG
GF
x H. Hence, there exists G F such that x G. Since, G is an open set Now if x G, y G, then either Ix = Iy or Ix Iy = . For if Ix 6= Iy
there exists r > 0 such that B(x; r) G. But G H. Hence, B(x; r) H. then Ix Iy would be an open interval contained in G containing x which is
Hence, H is an open set. larger than Ix . This contradicts definition of Ix . Finally, each Ix contains a
rational number. Since disjoint intervals cannot contain the same rational
Theorem 3.4 Every subset of (R, d) is an open set. number and since there are only countably many rational numbers, there
Proof. Note that all one point subsets of (R, d) are open sets. Hence, using cannot be uncountably many mutually disjoint intervals Ix . Hence, every
theorem (3.3), we get the required result. open set G R can be written as a finite number or a countable number
of open intervals which are mutually disjoint.
Theorem 3.5 If G1 and G2 are open subsets of M then G1 G2 is also
open. Example 3.2 Show that arbitrary intersection of open sets need not be
Proof. We may assume that G1 G2 6= . Let x G1 G2 . Hence, x G1 an open set.
and x G2 . Since, G1 and G2 are open sets there exist positive real numbers Solution 1. Let Gn = B(0, n1 ), n N. Then each Gn is an open set in R
r1 and r2 such that B(x; r1 ) G1 and B(x; r2 ) G2 . If r = min{r1 , r2 } as each Gn is an open ball and Gn = {0} which is not an open set in R.
then B(x; r) G1 and B(x; r) G2 and thus B(x; r) G1 G2 . Hence,
G1 G2 is an open set. Or
Corollary 3.1 The intersection of finite number of open sets is open. Solution 2. Let Gn = B( 12 , 12 + n1 ), n N. Then each Gn is an open set
in R as each Gn is an open ball and Gn = [0, 1] which is not an open set
Theorem 3.6 Every open set G R can be written as G = In where in R.
I1 , I2 , . . . are a finite number or a countable number of open intervals which 1 Note that an open ball in R is an open interval.
are mutually disjoint. (That is Im In = if m 6= n.) 2 We do not discard the possibility that Ix may be an unbounded interval.
3.3 Closed sets Definition 3.6 Let E be a subset of the metric space M. A point x M
is called a limit point of E if there is a sequence {xn } of points of E which
Definition 3.4 Let F be a subset of the metric space M. We say that F converges to x. The set E of all limit points of E is called the closure of E.
is a closed subset of M if F c is an open set.
Remark 3.3 Every point x of E is a limit point of E. For the sequence
Theorem 3.7 If F1 and F2 are closed subsets of M then F1 F2 is also x, x, . . . converges to x. Thus, if x E then x E, that is E E.
closed.
Remark 3.4 Note that every cluster point of E is a limit point of E.
Proof. Note that (F1 F2 )c = F1c F2c . Now F1c F2c is an open set as
However, the converse does not hold.
F1c , F2c are open and finite intersection of open sets is an open set.
Example 3.5 Let E = (0, 1). Note that 0 and 1 are limits points of E.
Using induction, we get However, 0
/ E and 1
/ E. Hence, E is a proper subset of E.
Corollary 3.2 Union of finitely many closed sets is closed.
3.4 Sequences in a metric space
Note that arbitrary union of closed sets need not be closed.

1 Definition 3.7 Let (M, d) be a metric space. A function s with domain
Example 3.3 Consider Fn = . Since, Fn is singleton and singletons N and having values in M is called an infinite sequence in M. The value
n

[ of s at n, namely s(n), is called the nth term of the sequence and is often
are closed each Fn is a closed set. However, E = Fn is not a closed set denoted as sn . Also, then the sequence is denoted as {sn }nN or simply as
n=1 {sn }. The set of all terms of a sequence, namely {sn |n N}, is called the
as 0 is a limit point of E. range of the sequence {sn }. If all the terms sn are real numbers, then {sn }
is called a real sequence.
T 3.8 If F be any family of closed subsets of a metric space M,
Theorem
then F is also closed. Definition 3.8 Let (M, d) be a metric space and {sn }
n=1 be a sequence
F F of points in M. We say that sn approaches L M as n approaches infinity
F )c =
S c
F . Since, F c is an open set and the
T
Proof. Note that ( if > 0 there is a positive integer r such that
F F F F
arbitrary union of open sets is an open set, the result follows.
d(sn , L) < , for every n r. (3.1)
Example 3.4 Give an example of a metric space (X, d) such that every In this case we write
proper subset of X is both open and closed.
Solution. In discrete metric space (X, d), singleton sets are open as if lim sn = l or lim sn = l or n L as n
n
x X then B(x, 12 ) = {x}. Since arbitrary union of open sets is open,
and we say that {sn }
n=1 is convergent in M to the point L.
every set is open. Thus, if E X then E is open. But E c is open as well.
Thus, E is closed as well as open. Thus, every proper subset of X is both Theorem 3.9 If the sequence {sn }
n=1 is convergent, then its limit is
open and closed. unique.
Definition 3.5 Let E be a subset of the metric space M. A point a M Proof of this theorem is left as an exercise.
is called a cluster point of E if for every h > 0 there exists a point x of E Definition 3.9 Let (M, d) be a metric space and {sn }
n=1 be a sequence
such that 0 < d(a, x) < h. of points in M. We say that sn is a Cauchy sequence if > 0 there is a
positive integer r such that
Remark 3.2 In other words, a M is said to be a cluster point of E if
every open ball B(a; h) contains a point of E distinct from a. d(sn , sm ) < , for all m, n r. (3.2)
Theorem 3.10 Let (M, d) be a metric space and {sn }n=1 be a sequence Proof. Since, E E we need to prove that E E. Let x E. To show
of points in M. Then {sn }
n=1 is a Cauchy sequence. that x E, it suffices to show that any open ball B[x; r] contains a point
of E. x E, the ball B[x; r] contains a point y E. Let d(x, y) = s.
The converse of this theorem is not true. There are metric spaces in which
Choose a positive number t such that t < r s. Since, y E, the ball
Cauchy sequences need not be convergent. For example, let M = (0, 1)
1 B[y; t] contains a point z E. Now,
with usual metric and sn = . It is easy to see that {sn } is a Cauchy
n+1 d(x, z) d(x, y) + d(y, z) < s + t < s + (r s) = r.
sequence but it is not convergent.
Hence, z B[x; r]. Thus, every open ball B[x; r] contains a point of E.
Definition 3.10 Let (M, d) be a metric space. We say that (M, d) is a
Hence, x is a limit point of E, that is, x E.
Complete metric space if every Cauchy sequence in M is convergent in M.
Theorem 3.13 In any metric space (M, d) the sets M and are both
Example 3.6 If {sn } is a Cauchy sequence in (R, d) then {sn } is eventually closed.
constant and hence convergent. That is, there exist a positive integer M Proof. Since the set M contains all its limit points and that has no limit
such that for all n M, xn = xn+1 = . . . . points (and hence contains all its limit points), both M and are closed.
1
If {sn } is a Cauchy sequence in (R, d) then for = there exist a
2 Theorem 3.14 Let G be an open subset of the metric space M. Then
1 G = M G is closed. Conversely, if F is a closed subset of M, then
positive integer M such that d(sn , sm ) < for all n, m M. Hence,
2 F = M F is open.
d(sn , sm ) = 0 for all n, m M, that is, for all n M, sn = sn+1 = . . . .
Proof. Let G be an open subset of the metric space M. If x G, then
In fact, if (M, d) is a discrete metric space, then every Cauchy sequence
there exists r > 0 such that B[x; r] G. Hence, B[x; r] contains no point
is constant. Thus, every discrete metric space is a complete metric space.
of G . Hence, x cannot be a limit point of G . Thus, no point of G is a limit
Theorem 3.11 Let E be a subset of the metric space M. Then the point point of G , so G contains all its limit points and is thus closed.
x M is a limit point of E if and only if every open ball B[x; r] about x Now, suppose F is a closed subset of M. If y F then as F is closed y
contains at least one point of E. is not a limit point of F. Hence, there exist r > 0 such that B[y; r] contains
Proof. Suppose x is a limit point of E. Then there is a sequence {xn } n=1
no point of F. That is, B[y; r] F . Hence, F is open.
of points of E that converges to x. If B[x; r] is any open ball about x, then
Example 3.8 Note that Z is a closed subset of (R, d) as Z =
S
(n, n+ 1)
there exists M N such that for all n M, d(xn , x) < . Hence, B[x; r] nZ
contains a point of E. and (n, n + 1) is an open set and hence Z is open. Hence, Z is a closed set.
Conversely, let x M and suppose every open ball B[x; r] about x
Solution 2. An alternate way to do this, is to show that Z has no cluster
contains at least one point of E. Then for every n N, the open ball
1 point. If x R Z is a cluster point then we write x = [x] + , 0 < < 1.
B[x; 1/n] contains a point xn E. Since, d(xn , x) < for all n N, the Let r = 12 min{, 1 }. Now, B(x, r) contains no point of Z. Hence, x is
n
sequence {xn }n=1 converges to x. Hence, x is a limit point of E. not a cluster point of Z. Hence Z is a closed set. 3
Example 3.7 Let (M, d) be a metric space and x M. Then {x} is a Definition 3.11 Let M be a metric space. The subset A of M is said to
closed subset of M. For the only sequence of points of {x} is x, x, x, . . . , be dense in M if A = M.
and hence x itself is the only limit point of {x}. Thus, {x} contains all its
Remark 3.5 1. We note that R {0} is an open set of R which is also
limit points and is therefore closed.
dense. In fact, we can remove finitely many points from R and the
Theorem 3.12 If E is any subset of a metric space M, then E is closed. resulting set is dense. It is easy to see that R Z is dense in R.
That is E = E. 3 Equivalently, one may show that every Cauchy sequence in Z is convergent.
2. Note that as every irrational number is limit of sequence of rationals, Exercise 3.6 1. Show that in a metric space, the complement of a sin-
Q is dense in R. (usual metric) gleton set is open.
3. We note that Qc = R Q (also) is dense in R. 2. Prove that in a metric space, the complement of a finite set is open.
4. A is dense in M if every point in M is a limit point of A. 3. Find the closure of the following subsets of R with usual metric
(a) A = {1/x|x N} (b) Set of integers (c) Q
Example 3.9 Note that the discrete metric space (R, d) has no dense sub- (d) (0, 1) (e) (0, 1] (f) [0, 1).
set except R itself as every subset of (R, d) is both open and closed. Thus, 4. Give an example of a metric space in which
if A R then A = A. Thus, if A 6= R then A cannot be dense in (R, d). (a) every set is both open and closed
(b) to show that intersection of an infinite number of open sets need
Example 3.10 Let S = {n + m 2|a, b Z}. Let a, b R such that a < b. not be open
Then there exists an s S such that a < s < b. Thus, S is dense in R.
(c) of an infinite class of closed sets whose union is not closed
Solution.
If x, y S and k Z then x y, kx S.4 If n + m 2 = (d) of a set which is neither closed nor open

n + m 2 then n = n and m = m . 5 For, if m 6= m then 2 = mnn m , a (e) to show that the intersection of an infinite number of neighbour-
contradiction. hood of a set, need not be a neighbourhood of that set.
Let Sm = m 2 [m 2]. Thus, 0 Sm < 1.
Claim 1. If m 6= m then Sm = Sm
. For if, Sm = Sm then Solutions

m 2 [m 2] = m 2 [m 2] then [m 2] + m 2 = [m 2] + m 2. 1. Let x0 M. If y 6= x0 then r = d(x0 , y) > 0. Note that x0
/ B(y, r/2).
Hence, {x0 }c is open.
Thus, m = m .
2. Follows from 1 and from the fact that finite intersection of open sets
Hence, we conclude that {Sm |m Z} is an infinite subset of S [0, 1).
is open.
Given > 0, we partition [0, 1) into k sub-intervals so that each sub-
interval has length < . At least one of these sub-intervals must contain 3. (a) A {0} (b) Z (c) R (d), (e) and (f) [0, 1].
two distinct elements, say Sm and Sm . Without loss of generality, we may
assume that Sm < Sm . Thus, 0 < Sm Sm < .
Claim 2. Let > 0 such that < b a. Then there exists an n Z such
that a < n < b.
We may assume that 0 < a < b. By Archimedean property there exists
N such that a < N . Using well ordering principle, choose n to be the least
positive integer k such that k > a. Thus, (n 1) < a n. We claim that
n < b. For otherwise, b a n (n 1) = , a contradiction.
ba
We take = . Then there exists s S such that 0 < s < . Hence
2
there exists an integer n such that a < ns < b. 6 Since, ns S, the claim
is proved.
4 Note that S is an integral domain. It is also true
5 This
that S is a Euclidean domain.
means that S is a free Z-module and {1, 2} is a basis.
6 Take s = as in claim 2.
4.1 Continuous functions

Definition 4.2 Let (M1 , d1 ) and (M2 , d2 ) be metric spaces and f be a
Chapter 4 function from M1 to M2 and a M1 . We say that f is continuous at a if
and only if for every sequence (xn ) converging to a, we have the sequence
(f (xn )) converging to f (a).
Limits in Metric spaces
Theorem 4.2 Let (M1 , d1 ), (M2 , d2 ) and (M3 , d3 ) be metric spaces and
Definition 4.1 Let (M1 , d1 ) and (M2 , d2 ) be metric spaces. Let E be f be a function from M1 to M2 and g be a function from M2 to M3 . If
a subset of the metric space M and a be a cluster point of E. Let f : f is continuous at a M1 and g is continuous at f (a) M2 , then g f is
E {a} M2 be a function. We say that f (x) approaches L( M2 ) continuous at a.
as x approaches L if given > 0 there exists > 0 such that whenever Proof. If {xn }n=1 is a sequence of points in M1 converging to a, then it
0 < d1 (x, a) < and x E, d2 (f (x), L) < . suffices to show that

lim g[f (xn )] = g[f (a)].
n
In this case, we write
lim f (x) = L. Since, f is continuous f (xn ) converges to f (a). Also, g is continuous at
xa f (a). Hence, g(f (xn )) converges to g(f (a)). This proves the theorem.
Remark 4.1 It is easy to see that if lim f (x) exists then it is unique. For Theorem 4.3 Let (M, d) be a metric space and a M. Let f and g be real
xa
if, L and L are the limits then given > 0 there exist 1 > 0 such that valued functions continuous at a. Then f + g, f g and f g are continuous
whenever 0 < d1 (x, a) < 1 and x E, d2 (f (x), L) < /2 and there exist at a. Further, if g(a) 6= 0, then f /g is continuous at a.
2 > 0 such that whenever 0 < d1 (x, a) < 2 and x E, d2 (f (x), L ) < /2. Proof. If {xn } n=1 is a sequence of points in M1 converging to a, then
If x E such that d1 (x, a) min{1 , 2 } then d2 (L, L ) d2 (L, f (x)) + {f (xn } and {g(xn )} converge to f (a) and g(a) respectively. Using algebra of
d2 (f (x), L ) < . Hence, d(L, L ) = 0 which implies that L = L . limits of sequences, it follows that {f (xn )+ g(xn )} converges to f (a)+ g(a),
i.e {(f + g)(xn )} converges to (f + g)(a). Hence, it follows that f + g is
We have studied algebra of limits for real valued functions of a real variable. continuous at a. Similarly, f g, f g and f /g are continuous at a.
The same results hold in a general case. We state the algebra of limits and
leave the proof as an exercise. Definition 4.3 Let (M1 , d1 ) and (M2 , d2 ) be metric spaces and f be a
function from M1 to M2 . We say that f is continuous from M1 into M2 (or
Theorem 4.1 Let (M, d) be a metric space and let a be a cluster point of simply f is continuous on M1 ) if f is continuous at each point in M1 .
M. Let f and g be real valued functions from M to R. If lim f (x) = L and
xa
lim g(x) = N, then Theorem 4.4 Let (M, d) be a metric space and a M. Let f and g be real
xa
valued functions continuous on M1 . Then f + g, f g and f g are continuous
1. lim (f (x) + g(x) = L + N. on M1 . Further, if g(x) 6= 0 for every x M1 then fg is continuous on M1 .
xa
2. lim (f (x) g(x)) = L N. Example 4.1 Note that f (x) = x2 is a continuous function from R onto
xa
[0, ) while g(x) = sin(x) is a continuous function from R onto [1, 1].
3. lim f (x)g(x) = LN.
xa 1
Theorem 4.5 Every function from (R, d) into a metric space is contin-
f (x) L uous on (R, d).
4. lim = , provided N 6= 0.
xa g(x) N 1 discrete metric
25
(
Proof. If {xn }n=1 is a sequence of points in (R, d) converging to a, then x if x is rational,
{xn } is a Cauchy sequence. Hence, using Example 3.6, there exist M N Example 4.3 Let f : R R defined by f (x) =
1 x if x is irrational.
such that xn = xn+1 = xn+2 = . . . for all n M, i.e., {xn } is a constant 1
sequence after the M th term. Hence, {f (xn )} is a constant sequence after We will show that f is discontinuous at every point a R except a = .
2
the M th term and hence convergent. Hence, f is continuous. Let a be an irrational number and {xn } be a sequences of rational numbers
( which converges to a. Note that f (a) = 1 a and f (xn ) = xn and hence
1 if x is rational, {f (xn )} will converge to a 6= f (a). Hence, f is discontinuous at every
Example 4.2 Let : R R defined by (x) =
0 if x is irrational. irrational number a R.
We will show that is discontinuous at every point a R. Let {xn } and Let b be a rational number and {xn } be a sequences of irrational num-
{yn } be sequences of rational and irrational numbers respectively which bers which converges to b. Note that f (b) = b and f (xn ) = 1 xn and
converge to a. Note that {(xn )} will converge to 1 and {(yn )} will con- hence {f (xn )} will converge to 1 b. Now, b = 1 b if and only if b = 12 .
verge to 0. Hence, is discontinuous at every a R. Hence, f is discontinuous at every rational number except 12 .
Now we discuss the continuity of f at 12 . Note that
4.2 Equivalent Definitions of Continuity
f (x) f ( 1 ) = x 1

2 2
Theorem 4.6 Let (X, d1 ) and (Y, d2 ) be metric spaces and f be a function
from X to Y and x X.Then the following are equivalent: Hence, given > 0, take = . Now, |f (x)f ( 21 )| < whenever |x 12 | < .
1
1. f is continuous at x. Hence, f is continuous only at .
2
2. Given > 0 there exists > 0 such that d2 (f (x), f (a)) < whenever Example 4.4 2
Let f : R+ R defined by
d(x, a) < .
0 if x is irrational,
3. Given an open set V containing f (x) Y, we can find an open set U

1

containing x such that f (U ) V. f (x) = if x is a rational number of the form m
n , where

n
m, n > 0 are integers and gcd(m, n) = 1.

Proof. Suppose (2) holds i.e. given > 0 there exist > 0 such that
d2 (f (x), f (a)) < whenever d1 (x, a) < . If {xn }
n=1 is a sequence of points Claim: f is continuous at every irrational number and discontinuous at
in X converging to a, then corresponding to > 0 there exist M N such every rational number.
that d1 (xn , a) < for all n M. Hence, d2 (f (xn ), f (a)) < whenever Let a > 0 be a rational number and {xn } be a sequence of irrational
n M. Hence, the sequence {f (xn )} n=1 of points in M2 converges to f (a). numbers which converges to a. Now {f (xn )} will converge to 0 while f (a) >
Conversely, assume that (1) holds. Suppose (2) does not hold. Then 0. Hence, f is discontinuous at a. Thus, f is discontinuous at every rational
there exists > 0 such that for every > 0 there exists x M1 such that number.
1
d1 (x, a) < but d2 (f (x), f (a)) . In particular, for = there exists On the other hand, if b is an irrational number and > 0 then by the
n 1
xn M1 such that d1 (xn , a) < but d2 (f (xn ), f (a)) . Thus, xn a as Archimedean property, there is a natural number n0 such that < .
n0
n but f (xn ) does not tend to f (a), a contradiction. Hence, (2) holds. There are only a finite number of rationals with denominator less than
(1) if and only if (3) is left as an exercise. n0 in the interval (b 1, b + 1). Hence, > 0 can be chosen so small
that the neighbourhood (b , b + ) contains no rational numbers with
Also, one can show that (1) (2) (3) (1). We leave this as an
exercise. 2 This function was introduced by K.J.Thomae in 1875.
denominator less than n0 . It then follows that for |x b| < , x R+ , 2. The set G M1 is open if and only if its image f (G) M2 is open.
1
we have |f (x) f (b)| = |f (x)| < . Hence f is continuous at every 3. The set F M1 is closed if and only if its image f (G) M2 is closed.
n0
irrational number.
Thus, f is continuous at every irrational number and discontinuous at Proof. Follows using Theorem 4.7 and Theorem 4.8.
every rational number. Hence, f is Riemann integrable over [0, 1].
Example 4.5 Let X, Y be metric spaces and f, g : X Y be continuous
Exercise Set functions. Let E = {x|f (x) 6= g(x)}. Show that E is an open set.
Solution. Let x E. Hence, f (x) 6= g(x). Thus, = d(f (x), g(x)) > 0.
1. Give an example of a function which is continuous on R and whose Let B1 = B(f (x); /2) and B2 = B(g(x); /2). Thus, B1 B2 = . Since,
range is (0, ) f and g are continuous functions, there exist 1 > 0 and 2 > 0 such that
2. Show that every function from Z into a metric space is continuous. f (B(x; 1 )) B1 and g(B(x; 2 )) B2 . Thus, if = min{1 , 2 }, then
f (B(x; )) B1 and g(B(x; )) B2 . Since, B1 B2 = we get that
3. Let (X, d) be a metric space. If is the product metric defined on B(x; ) E. Hence, E is an open set.
X X by ((x1 , y1 ), (x2 , y2 )) = max{d1 (x1 , x2 ), d2 (y1 , y2 )} show that
d is a continuous function from X X to R. Or
Theorem 4.7 Let (M1 , d1 ) and (M2 , d2 ) be metric spaces and f : M1 Let x be a limit point of E c and (xn ) be a sequence of points in E c con-
M2 be a function. Then f is continuous on M1 if and only if f 1 (G) is an verging to x. Since, f and g are continuous functions, f (xn ) and g(xn )
open subset of M1 whenever G is an open subset of M2 . converges to f (x) and g(x). Since xn E c , f (xn ) = g(xn ) for every n.
Proof. Suppose f is continuous on M1 . We wish to show that if G Thus, f (x) = g(x) and E c is a closed set. Hence, E is an open set.
is open in M2 then f 1 (G) is an open subset of M1 . If x f 1 (G)
then y = f (x) G. Since G is an open set there exist s > 0 such Example 4.6 Show that one can construct a continuous function from a
that B[y; s] G. Since, f is continuous at x, corresponding to this s, metric space (M1 , d1 ) to a metric space (M2 , d2 ) which maps closed sets to
there exists r > 0 such that d1 (x1 , x) < r implies d2 (f (x1 ), f (x)) < s, closed sets but does not map open sets to open sets.
that is, f (B[x; r]) B[y; s] or equivalently, f 1 (B[y; s]) B[x; r]. Thus, Solution. This is easy. For example, consider a constant function from R
B[x; r] f 1 (B[y; s]) f 1 (G). Hence, f 1 (G) is an open set. to R.
Conversely, let a M1 and b = f (a). Let s > 0 and B = B[a; s].
Example 4.7 Show that there exist a continuous function from a metric
Now B is an open set. Hence, f 1 (B) is an open set in M1 . Further,
space (M1 , d1 ) to a metric space (M2 , d2 ) which maps open sets to open
a f 1 (B). Since, f 1 (B) is an open set in M1 there exists r > 0 such
sets but does not map closed sets to closed sets.
that B[a; r] f 1 (B). Thus, f is continuous. 1
Solution. Consider f : R+ R where f (x) = . Note that f is one-one,
Theorem 4.8 Let (M1 , d1 ) and (M2 , d2 ) be metric spaces and f : M1 x
continuous function. It is easy to see that f maps open sets to open sets.
M2 be a function. Then f is continuous on M1 if and only if f 1 (F ) is a However f (N) is not a closed set in R as 0 is a cluster point even though N
closed subset of M1 whenever F is a closed subset of M2 . is a closed set in R+ .
Proof. Follows from Theorem 4.7 and Theorem 3.14
Definition 4.4 Let (M1 , d1 ) and (M2 , d2 ) be metric spaces and f : M1
Theorem 4.9 Let (M1 , d1 ) and (M2 , d2 ) be metric spaces and f : M1
M2 be a one-one and onto function. If both f and f 1 are continuous (on
M2 be a one-one and onto function. Then if f has one of the following
M1 and M2 , respectively) then we call f a homeomorphism from M1 onto
properties then it has them all.
M2 . If a homeomorphism from M1 onto M2 exists, then we say that M1
1. Both f and f 1 are continuous (on M1 and M2 , respectively). and M2 are homeomorphic.
Remark 4.2 Theorem (4.9) tells us that if f : M1 M2 be a one-one and is a one-one function. Further f g = IdJ and g f = IdI . Hence, f and
onto function such that G is open in M1 if and only if its image f (G) M2 g are onto functions and are inverses of each other. Hence, I and J are
is open then f is a homeomorphism. Equivalently, if f : M1 M2 be a homeomorphic.
one-one and onto function such that F is closed in M1 if and only if its
Example 4.13
image f (F ) M2 is closed then f is a homeomorphism.
Example 4.8 Note that the metric spaces [0, 1] and [0, 2] with absolute
value metric are homeomorphic. For if f (x) = 2x, then f is a homeomor- Consider the function f : (1, 1) R
phism from [0, 1] onto [0, 2]. x
defined by f (x) = . Note that
1 x2
f (x) is a rational function i.e. it is
)
Example 4.9 Note that the metric spaces (0, 1) and (0, with absolute
a ratio of two polynomials. Hence, it
value metric are homeomorphic. For if f (x) = tan x , then f is a
2 is continuous at all points where de-
homeomorphism from (0, 1) onto (0, ). nominator does not vanish. In this
case, the denominator never vanishes as
Example 4.10 Note that the metric spaces R and (0, ) with absolute x (1, 1). Further, it is easy to see
value metric are homeomorphic. For if f (x) = ex , then f is a homeomor- that f (x) is one-one and onto function.
phism from R onto (0, ). Can you determine the inverse of f ? Is
f 1 continuous?
Example 4.11 Note that the metric spaces (0, ] and R with absolute
value metric are homeomorphic. For if g(x) = log x, then g is a homeomor- x
phism from (0, ) onto R. Graph of y =
1 x2
Exercise 4.1 1. Show that (0, 1) and R are homeomorphic.
Example 4.12 Show that any two closed and bounded intervals in R are
homeomorphic. 2. Show that R+ and R are homeomorphic.
Let I = [a, b] and J = [c, d]. There are two possibilities:
Case I: a = c. Thus, I = [a, b] and J = [a, d]. Define f : I J by 3. Show that (0, 1] and [1, ) are homeomorphic.
f (x) = a + (d a)/(b a)(x a). Note that f is a polynomial. Hence, 4. Show that R2 and C are homeomorphic.
f is continuous. Further, f (x) = f (y) a + (d a)/(b a)(x a) =
a + (d a)/(b a)(y a) x = y. Thus, f is a one-one function. Define Solutions
g : J I by g(x) = a + (b a)/(d a)(x a). Note that g is a polynomial. 1. Define f : (0, 1) R by f (x) = x 12 . f maps (0, 1) to ( 12 , 12 ).
Hence, g is continuous. Further, g(x) = g(y) a + (b a)/(d a)(x a) = Define g : ( 12 , 12 ) R by g(y) = tan(y). Note that f and g are
a + (b a)/(d a)(y a) x = y. Thus, g is a one-one function. Further homeomophisms. By taking their composition, we get the function.
f g = IdJ and g f = IdI . Hence, f and g are onto functions and are
inverses of each other. Hence, I and J are homeomorphic. 2. Note that f (x) = log(x) is a one-one onto continuous function from
R+ to R and the inverse map is x 7 ex .
Case II: a 6= c. Define f : I J by f (x) = a + (d b)/(c a)(x a). Note
that f is a polynomial. Hence, f is continuous. Further, f (x) = f (y) 1
3. Note that f (x) = is a one-one onto continuous function from (0, 1]
a + (d b)/(c a)(x a) = a + (d b)/(c a)(y a) x = y. Thus, f is a x
one-one function. Define g : J I by g(x) = c+ (b d)/(a c)(x c). Note to [1, ) and the inverse map is f 1 (x) = x1 .
that g is a polynomial. Hence, g is continuous. Further, g(x) = g(y) 4. Define f : R2 C by f (x, y) = (x, y). This is a homeomorphism.
c + (b d)/(a c)(x c) = c + (b d)/(a c)(y c) x = y. Thus, g
Example 5.1 Note that every subset of a bounded set is bounded. Also,
every subset of (R, d) is a bounded set as d(x, y) 1 for all x, y R.
Chapter 5 Definition 5.5 Let (M, d) be a metric space. We say that the subset A
of M is totally bounded if, given > 0, there exists a finite number of
subsets A1 , . . . , An of M such that diam Ak < , 1 k n and such that
Compact Metric Spaces A
[n
Ak .
k=1
5.1 Compact Set Example 5.2 If A is a totally bounded subset of the metric space (M, d)
then every subset of A is totally bounded.
Definition 5.1 Let M be any set, the[family F of subsets of A of M is Example 5.3 Show that every finite subset of a metric space (M, d) is
said to form a covering of M if M A.
totally bounded.
AF
Let A = {a1 , . . . , an }. Then Ai = B[ai , /3] cover A and diam Ai < .
Definition 5.2 Let (M, d) be a metric space and A be a subset of M. Hence, A is totally bounded.
The family
[ F of {Ui |i I} of M is said to form an open covering of A if Theorem 5.1 If the subset A of metric space (M, d) is totally bounded,
A Ui and each Ui F is an open set. then A is bounded.
iI
[ Proof. If A is totally bounded, then there exist nonempty subsets of M, say
n
If J I such that A Ui , we then say that {Ui |i J} is a subcover [
iJ
A1 , . . . , An , such that diam Ak < 1, 1 k n and such that A Ak .
of the given open cover of A. k=1
S
If there exists a finite subset J I such that A iJ Ui , then we say For every k, 1 k n, let ak Ak and D = d(a1 , a2 ) + d(a2 , a3 ) + +
that the given cover admits a finite subcover. d(an1 , an ). Let x, y A. Since, Ai s cover A there exist Ai and Aj such
that x Ai and y Aj for some i and j. We may assume that i j. Now,
Definition 5.3 Let (M, d) be a metric space and A be a subset of M. A d(x, y) d(x, ai ) + [d(ai , ai+1 ) + + d(aj1 , ai )] + d(aj , y)
is said to be compact if every open cover has a finite subcover.
d(x, ai ) + D + d(aj , y).
Since, diam Ai < 1 and diam Aj < 1, d(x, ai ) < 1 and d(aj , y) < 1. Hence,
5.2 Bounded and Totally Bounded sets
d(x, y) < 1 + D + 1 = D + 2, for all x, y A.
Definition 5.4 Let (M, d) be a metric space. We say that the subset A of
M is bounded if there exists a positive number L such that Hence, A is a bounded set.
Theorem 5.2 Every bounded subset of (R, d) is totally bounded.
d(x, y) L x, y A. Proof. Let A be a bounded subset of R. Then A [L, L] for some L > 0.
Given > 0, by Archimedean property, there exist an integer n such that
If A is bounded we define the diameter of A, denoted diam(A), as
n( ) > 2L and A is covered by
2
diam A = sup d(x, y).
x,yA [L, L + ], [L + , L + 2 ], . . . , [L + (n 1) , L + n ]
2 2 2 2 2
If A is not bounded we write diam(A) = . Hence, A is a totally bounded set.
33
n
Example 5.4 If we take M = [0, 1] with usual metric, then we note that [
that A Ak . For every k, 1 k n, let ak Ak . Then {a1 , . . . , an }
every subset of M is totally bounded.
k=1
is -dense in A. Hence, if A is totally bounded then A has a finite -dense
Theorem 5.3 Every bounded subset of (Rn , d) is totally bounded.
subset.
Proof. We will write the proof for n = 2. Let A be a bounded subset
Conversely, if {x1 , . . . , xn } is 3 -dense in A, then {B(x1 , 3 ), . . . , B(xn , 3 )}
of R2 . Then A [L, L] [L, L] for some L > 0. Given > 0, by
forms a covering of A by sets of diameter < . Hence, A is totally bounded.
Archimedean property, there exist an integer n such that n( ) > 2L. We
3
divide [L, L] [L, L] into small squares of length /3 by dividing [L, L] Theorem 5.6 Let (M, d) be a metric space. The subset A of the metric
on x-axis as well as y-axis as space (M, d) is totally bounded if and only if every sequence of points of A
contains a Cauchy subsequence.

[L, L + ], [L + , L + 2 ], . . . , [L + (n 1) , L + n ] Proof. Suppose A is totally bounded and (xn ) n=1 be a sequence of points
3 3 3 3 3 of A. We want to show that (xn ) has a Cauchy subsequence. Since the
n=1
Thus, we get n2 squares whose diagonal (and hence diameter) is of length set A is totally bounded, A can be covered by a finite number of subsets of A
< . Hence, A is totally bounded. of diameter < 1. One of these sets, call it A1 , must contain xn for infinitely
In a similar way, we can write the proof in general case.1 many values of n. Choose an integer n1 such that xn1 A1 . Since, A1 A
and A is totally bounded, it follows that A1 is totally bounded and hence
Theorem 5.4 Every totally bounded subset of the discrete metric space 1
A1 can be covered by a finite number of subsets of A1 of diameter < .
(R, d) contains only finitely many elements. 2
One of the sets, say A2 must contain xn for infinitely many values of n.
Proof. If A is totally bounded subset of (R, d), then there exist nonempty
Choose an integer n2 > n1 such that xn2 A2 . Since, A2 A1 we also
subsets A1 , . . . , An of M such that diam Ak < 1, 1 k n and such that
[n have xn2 A1 . Continuing in this way we obtain, for every natural number
A Ak . Since, diam Ak < 1, Ak contains exactly one element. Hence, n, a subset An of An1 such that diam Ak < 1/k, and xnk Ak . Since,
k=1 xnk , xnk+1 , . . . all lie in Ak , and diam Ak < 1/k, it follows that (xnk ) is a
A has only finitely many elements. Cauchy subsequence of (xn ).
Conversely, suppose every sequence (xn ) n=1 of points of some subset
Definition 5.6 Let A be a subset of the metric space (M, d) and > 0. A of M contains a Cauchy sequence. We wish to show that A is totally
The subset B of A is said to be -dense in A if for every x A there exists bounded. Suppose the contrary. Then there exists > 0 such that A
y B such that d(x, y) < . contains no finite -dense subset. Thus, if x1 A then the set {x1 } is
not -dense in A, so there exist x2 A such that d(x1 , x2 ) . But
Thus, B is -dense in A if each point of A is within distance from some
the set {x1 , x2 } is not -dense in A. Hence, there exist x3 A such
point of B.
that d(x1 , x3 ) , d(x2 , x3 ) . Continuing in this way we can con-
Theorem 5.5 Let (M, d) be a metric space. The subset A of M is totally struct a sequence {xn } n=1 of points of A such that d(xj , xk ) for any
bounded if and only if for every > 0 there exists a finite subset {x1 , . . . , xn } j, k N, j 6= k. But then (xn ) n=1 has no Cauchy subsequence, which con-
of A which is -dense in A. tradicts our hypothesis. Hence, A must be totally bounded.

Proof. Let > 0. If A is totally bounded, then there exist nonempty
Theorem 5.7 If the metric space (M, d) is compact then (M, d) is both
subsets A1 , . . . , An of M such that diam Ak < , 1 k n and such
complete and totally bounded.
1 LetA be a bounded subset of Rn . Then A [L, L] [L, L] for some L > 0. Proof. Let (M, d) be a compact metric space. Given > 0, the family

We divide [L, L] on each axis into closed intervals of length
n+1
.Now complete the {B(x, )|x M } is an open cover of M. Since, M is compact this open
proof. cover has a finite subcover. Hence, M is totally bounded.
We now show that M is complete. Let (xn ) be a Cauchy sequence in We claim that lim xn = l. Let > 0 be given. Now there exist M N
n
M. Then for every k N there exists nk such that d(xn , xnk ) < k1 for all such that
n > nk . Let |xn xM | < /2 for n M
1
Uk = {x M |d(x, xnk ) > }.
k using (5.2).
Note that Uk is complement of B[xnk , k1 ]. Hence, Uk is open. Now xn / Uk
Example 5.5 Note that (R, d), where d is a discrete metric is complete
for n > nS k . Hence, no finite subcover of Uk s cover X. For, if they did,
metric spaces. In fact, every discrete metric space is a complete metric
say M = m i=1 Ui , we take n > max{n1 , . . . , nm }. Then xn
/ Uk for any k
space as every Cauchy sequence in a discrete metric space is eventually
with 1 kS m. This implies that {Uk } cannot cover M. Thus there exists
constant.
x M i=1 Ui . But then d(x, xnk ) < k1 . Hence, xnk x. Since, (xn ) is
a Cauchy sequence in M, we see that (xn ) also converges to x. Thus, X is 1. Are the following subspaces of R with the usual metric complete?
complete. Give reasons for your answer.
(1) The set of all positive reals. (2) The set of rationals.
5.3 Complete Metric spaces (3) [1, 2] {3, 4}. (4) The set of integers. (5) The set of irrationals.
Definition 5.7 Let (M, d) be a metric space. M is said to be a complete Example 5.6 Show that (0, 1) with usual metric is not a complete metric
metric space if every Cauchy sequence of points in M converges to a point space.
in M . 1
Let an = . We show that (an ) is a Cauchy sequence. Let > 0.
n
Theorem 5.8 R is a complete metric space. 1 1
Choose M N such that < . If m, n > M then d(am , an ) = | m n1 | <
Proof. Let (xn ) be a Cauchy sequence in R. Let us take > 0. By the M 2
1 1
definition of Cauchy sequence, there exits N () such that m + n < . Hence, (an ) is a Cauchy sequence. This sequence converges to
0 but 0
/ (0, 1). Thus, (0, 1) is not a complete metric space.
|xn xm | < for m N (), n N (). (5.1)
Theorem 5.9 If (M, d) is a complete metric space and A is a closed subset
In particular, of M then (A, d) is a complete metric space.
|xn xN () | < for n N (). (5.2) Proof. Let {xn } n=1 be a Cauchy sequence of points in (A, d). Since, A
M, {xn }n=1 is a Cauchy sequence of points in (M, d). But (M, d) is a
If we take = 1 then there exist N = N (1) such that
complete metric space. Hence, {xn } n=1 converges to some x M. But x is
|xn xN | < 1 for n N. (5.3) a limit point of A because x is the limit of a sequence of points in A. Since,
A is closed, x A. Hence, every Cauchy sequence in A converges to some
This implies that xn (xN 1, xN + 1) for all n N. Let point in A. Hence, (A, d) is a complete metric space.
S = {x R| there exist infinitely many n such that xn x}. Theorem 5.10 Let (M, d) be a complete metric space. For each n N,
let Fn be a closed bounded subset of M such that
Note that xN 1 S and hence S is non-empty. Further, xN + 1 is an
upper bound of S. If this is not true, then there exists y S such that F1 F2 . . . Fn Fn+1 . . . , (5.4)
y > xN + 1 and there are infinitely many n such that xn y. This implies and diam Fn 0 as n . (5.5)
that xn > xN + 1 for infinitely many n, a contradiction. Thus, S is a
non-empty subset of R and xN + 1 is an upper bound of S. Hence, by LUB
T
axiom, S has a least upper bound, say l. Then Fn contains precisely one point.
n=1
T
Proof. For each n N, let an Fn . Then by (5.4), By (5.10), there is precisely one point x An . Now, since F is a
covering of M, there is a set G F such that x G. But G is open as F
an , an+1 , . . . all lie in Fn . (5.6) is an open covering and so there exists r > 0 such that B(x, r) G. Since,
r > 0 there exists a natural number N such that N1 < r.
Given > 0 there exists by (5.5), an integer M N such that Since, x AN and diam (AN ) < N1 < r, we get AN B(x, r) G.
diam (FM ) < . Hence, G alone covers AN , a contradiction. This proves the theorem.
Theorem 5.12 The metric space (M, d) is compact if and only if every
Now, aM , aM+1 , . . . all lie in FM . For n, m M, we have
sequence of points in M has a subsequence converging to a point in M.
d(an , am ) diam (FM ) < . Proof. Let (M, d) be compact metric space and that {xn } n=1 be a sequence
of points in M. Since, M is totally bounded, the sequence {xn } n=1 has a
This proves that {an } is a Cauchy sequence in M. Since, (M, d) is a complete Cauchy subsequence {xnk } k=1 . But {xnk
}
k=1 converges to a point in M
metric space, {an } is convergent to some a M. By (5.6), a is a limit point since M is a complete metric space. Thus, if M is compact then every

T sequence of points in M has a subsequence converging to a point in M.
of the closed set Fn for each n. Hence, a Fn for each n. Hence, a Fn . Conversely assume that every sequence in M has a convergent subse-
n=1
If b M such that a 6= b then d(a, b) > diam Fn for n sufficiently large. quence. Then, M is totally bounded. To show that M is complete we must

T
T show that every Cauchy sequence {xn } n=1 in M converges to a point in M.
Hence, b cannot be in Fn . Hence, Fn contains precisely one point.
n=1 n=1 By assumption, {xn } n=1 has a subsequence {xnk }
n=1 which converges to a

point x M. Since, {xn }n=1 is a Cauchy sequence, given > 0 there exists
M N such that d(xm , xn ) < /2 for all m, n M. Since, {xnk } n=1 which
5.4 Equivalent conditions for Compact metric spaces
converges to a point x M there exists r N such that d(xnk , x) < /2
Theorem 5.11 If (M, d) is totally bounded and complete metric space, for all k r, (i.e. nk nr ). If p = max{nr , M } then np p nr and for
then M is compact. all k p, we have
Proof. Suppose the contrary. Then for some open covering F , no finite
d(xk , x) d(xk , xnp ) + d(xnp , x) < /2 + /2 = .
number of sets in F form a covering of M. Now M is totally bounded and
hence may be written as the union of a finite number of bounded subsets Hence, the Cauchy sequence {xn }n=1 in M converges to a point x M.
each of whose diameter is less than 1. But one of these subsets, say A1 , Hence, M is a complete metric space. Since, M is also totally bounded, M
cannot be covered by a finite number of sets in F . (Otherwise all of M is a compact metric space.
would be covered by a finite number of sets in F .) We now consider A1 .
Note that diam A1 = diam A1 and A1 is a closed subset of M. Further, Corollary 5.1 If A is a closed subset of the compact metric space (M, d)
diam A1 < 1 and A1 cannot be covered by a finite number of sets in F . then the metric space (A, d) is also compact.
Since A1 is itself totally bounded, the same reasoning shows the existence Proof. Let {xn }
n=1 be a sequence of points in A. Since, A M, {xn }n=1 is
of a subset A2 whose diameter is less than 1/2 and which cannot be covered a sequence of points in M. Since, (M, d) is a compact metric space, {xn }
n=1
by a finite number of sets in F . Thus, A2 A1 , diam A2 < 1/2 and A2 has a subsequence converging to a point x M. This implies that x is a
cannot be covered by a finite number of sets in F . Continuing in this way, limit point of A. Since, A is closed, x A. Thus, every sequence in A has
for every n N, we can show the existence of An M such that a subsequence converging to a point in A. Hence, the metric space (A, d) is
also compact.
A1 A2 A2 , diam An < 1/n
Theorem 5.13 Let A be a subset of the metric space (M, d). If (A, d) is
and such that An cannot be covered by a finite number of sets in F . compact then A is a closed subset of the metric space (M, d).
Proof. Let x be a limit point of A. Then there is a sequence sequence Conversely, assume that whenever F is\ a family of closed subsets of M
{xn }
n=1 of points of A converging to x. But then {xn }n=1 is a Cauchy with the finite intersection property, then F 6= . We want to show that

sequence in A, and so, since A is complete, {xn }n=1 converges to a point in F F
A. This point must be x (as the limit of a sequence is unique, if exists) and M is compact. Consider G to be an open cover of M. If G does not have
so x A. Thus A contain all its limit points and so A is closed.2 a finite subcover of M , then this means that whenever G1 , G2 , . . . , Gn are
elements of G, M 6 G1 G2 Gn . Hence, there exists x M such that
Definition 5.8 A family F of subsets of a set M is said to have the finite x / G1 G2 Gn , that is, x (G1 G2 Gn )c . If Fi denote (Gi )c
intersection property if the intersection of any finite number of sets in F is then Fi is a closed set as each Gi is an open set. If F denote the set of all
never empty. Gc where G G then F is a family of closed sets S having finite intersection
c
( GG G)c ,
S
property. However, as G is an open cover, M GG G, M
Thus, a family F of subsets of a set M is said to have the finite intersection i.e. = M c ( GG Gc ). As Gc = F, we get F F F = , a contradiction.
T T
property if whenever F1 , . . . , Fn F , then F1 F2 Fn 6= . Hence, G has a finite subcover of M. That is, M is compact.
Example 5.7 Note that the family of all closed intervals [1/n, 1/n] has
the finite intersection property. Also, the family of open intervals (0, 1/n) 5.5 Continuous functions on compact metric space
has the finite intersection property.
Theorem 5.15 Let f be a continuous function from the compact metric
Theorem 5.14 The metric space M is compact if and only if whenever F space M1 into the metric space M2 . Then the range f (M1 ) of f is also
is\a family of closed subsets of M with the finite intersection property, then compact.
F 6= . Proof 1. Let G be any open covering of f (M1 ). Since, f is continuous
F F and G is an open set, f 1 (G) is an open subset of M1 . The family of all
Proof. Suppose first that M is a compact metric space and that F is a such sets f 1 (G) for G G is therefore an open covering of M1 . Since M1
family of closed subsets of M with the finite intersection property. We want is compact,a finite number of these sets, say f 1 (G1 ), . . . , f 1 (Gn ), form
to show that \ a covering of M1 . But then G1 , . . . , Gn form a covering of f (M1 ), and so
F 6= . (5.7) f (M1 ) is compact.
F F Proof 2. Let {yn } n=1 be any sequence in f (M1 ). For every n N let xn

For each F F let G = F = M f, and let G be the family of all such M1 such that f (xn ) = yn . Since, M is compact {xn } n=1 has a subsequence
open sets G. xnk which converges to a point, say x of M. Since f is continuous on
If F1 , . . . , Fn F , then using de Morgans laws, we get M1 f (xnk ) converges to f (x), that is, f (yk ) converges to f (x). Thus, any
sequence {yn } n=1 in f (M1 ) has a convergent subsequence and hence f (M1 )
F1 F2 Fn = M (G1 G2 Gn ). (5.8) is compact.
Corollary 5.2 Let f be a continuous function from the compact metric

\ [
F =M G. (5.9)
F F GG space M1 into the metric space M2 . Then the range f (M1 ) of f is a bounded
subset of M2 .
By the hypothesis, the left side of equation (5.8) is not empty, and so
Proof. This follows from the fact that f (M1 ) is compact and compact sets
M (G1 G2 Gn ) 6= . Hence, no finite number of sets in G form a
are bounded.
covering of M. Since, M is compact, it follows that G itself is not a covering
of M. Hence, right side of equation (5.9) is not empty, which is what we Definition 5.9 Let f be a function from a set A into a metric space M.
wanted to show. We say that the function f is bounded if its range f (A) is a bounded subset
2 Note that the proof uses only the completeness of A. of M.
Corollary 5.3 Let f be a real valued function continuous on a closed and Proof. We want to show that f 1 is continuous. Let F be a closed subset
bounded interval in R. Then f must be bounded. of M1 . We want to show that the inverse image of F under f 1 is closed
Proof. Since, every closed and bounded interval is compact, we get the in M2 . But the inverse image of F under f 1 is f (F ). Since, M1 is
required result. compact and F is closed, F itself is compact. Since, f is continuous f (F ) is
compact and hence a closed subset of M2 = f (M1 ). Hence, f is continuous.
Theorem 5.16 Let f be a continuous function from the compact metric
space M into R. Then f attains a maximum value at some point of M. Example 5.9 Let g : [0, ) [0, ) defined by g(x) = x. We want to
Also, f attains a minimum value at some point of M. prove that g is a continuous function. Let N N. Then f : [0, N ] [0, N 2 ]
Proof 1. Note that f is a bounded function. Let = sup f (x) and = defined by f (x) = x2 is a continuous function from the compact metric
xM space to compact metric space. Hence, f is a homeomorphism. Hence, f 1
inf f (x). Hence, and are limit points of f (M ). Since, f (M ) is compact, is a continuous function on [0, N 2 ]. But f 1 is the restriction of g to [0, N 2 ].
xM
f (M ) is a closed subset of R. Hence, f (M ) and f (M ). Hence, there Hence, g is continuous.
exist a, b in M such that f (a) = and f (b) = .
Proof 2. With the notation as in Proof 1, suppose / f (M ), then the 5.7 Uniform Continuity
function g(x) = f (x) is a continuous function on M and g never takes
the value 0 on M. Hence, 1/g is continuous on M. Since, M is compact, Definition 5.10 Let (M1 , d1 ) and (M2 , d2 ) be metric spaces and f be a
1 1 function from M1 to M2 . We say that f is uniformly continuous on M1
1/g is bounded (being continuous). Thus, = N for some
g(x) L f (x) if given > 0 there exists > 0 such that d2 (f (x), f (a)) < whenever
1 1 d1 (x, a) < .
N 0. and all x M. Hence, f (x) L for all x M. Hence, L
N N
is an upper bound for f (M ), a contradiction. Hence, f (M ). Exercise 5.1
Similarly, we can prove that f (M ).
1. Show that if (M, d) is a discrete metric space then every subset is
Corollary 5.4 Let f be a real valued function continuous on a closed and both open and closed. Further, show that its every compact subset is
bounded interval in [a, b]. Then f attains a maximum and a minimum value finite.
at points of [a, b].
Proof. Since, every closed and bounded interval is compact, we get the 2. Show that every finite subset is compact.
required result. 3. Show that {B(x, 1/2)}xX is an open cover of X.
Example 5.8 Let C[0, 1] denote the collection of all real valued bounded 4. Show that {B(x, n)}nN is an open cover of X.
continuous functions defined on [0, 1]. Define the norm of f C[0, 1] by,
kf k = sup{|f (x)| : x [0, 1]} Then prove that the mapping d : C[0, 1] R, 5. Show that a closed subset of a compact metric space is compact.
defined by d(f, g) = kf gk = sup{|f (x) g(x)| : x [0, 1]} is a metric for
6. Show that a closed subset of a complete metric space is complete.
C[0, 1].
7. If (xn ) is a Cauchy sequence and it has a convergent subsequence
converging to x then show that (xn ) converges to x.
5.6 Continuity of the inverse function
8. Show that following sets are closed subset of R3 . Check whether each
Theorem 5.17 If f is a one-one continuous function from the compact of them is compact?
metric space M1 onto the metric space M2 , then f 1 is continuous on M1 ,
and hence f is a homeomorphism. (a) {(x, y, z)|x2 = y 2 + z 2 }
(b) {(x, y, z)|x2 y 2 + z 2 = 1}

(c) {(x, y, z)|x2 + y 2 + z 2 = 1}
(d) {(x, y, z)|x y + z = 1}
9. Show that there is a homeomorphism between circle x2 + y 2 = 1 and
ellipse x2 /a2 + y 2 /b2 = 1.
10. If f : X Y is a continuous function and K is a compact set in X
then show that f (K) is compact.
11. Show that every convergent is a Cauchy sequence. But the converse
is not true.
A B = . Thus, A and B are open sets. Thus, A and B are non-empty

proper subsets of M which are both open and closed, a contradiction as M
is connected. Therefore, f is a constant function.
Chapter 6 Conversely, let us assume that M is not connected. Therefore there
exist two disjoint proper non-empty subsets A and B in M such that A
Connected Spaces and B are both open and closed in M and A B = M. Now we define
f : M {1} by (
1 if x A
Definition 6.1 A metric space X is said to be connected if the only subsets f (x) =
1 if x B.
which are both open and closed in X are and the space X itself.
A subset A of a metric space X is said to be connected if A is a connected Then, f : M {1} is a non-constant function. It is easy to see that
space when considered as a metric space (with the metric induced from X). f is continuous function as the inverse image of every open (closed) set of
{1} is an open (closed) set. Note that all the open (closed) sets of {1}
Theorem 6.1 R is connected. are {1}, { 1}, {1}, and their inverse images are M, A, B, , respectively,
Proof. We need to show that if A is a non-empty subset of R which is both which are open (closed) . 1
open and closed in R then A = R. Since, A 6= , there is an x A. Since, A
is a non-empty subset of R there exists an > 0 such that (x , x+ ) A. Corollary 6.1 A subset A of a metric space (M, d) (with the metric in-
Let duced from M ) is connected if and only if every continuous function f :
S = {b |(x , b) A, x < b}. A {1} is a constant function.
Note that x + S. Hence, S 6= . If S is bounded above then by LUB
Theorem 6.3 Let (M, d) be a connected metric space and g : M X be
axiom, S has supremum, say . Let n N. Since, n1 < , there exists
a continuous function. Then g(M ) is connected.
an xn S such that n1 < xn < . Thus, (x , xn ) A.
Proof 1. Assume that g(M ) is not connected. Then there exists a non-
By sandwich principle, (xn ) . Hence, is a limit point of A. But
empty proper subset V of g(M ) which is both open and closed in g(M ).
A is closed, hence A. But A is open. Hence, there exists t > 0 such
Since g is a continuous function, g 1 (V ) and g 1 (g(M ) V ) are both non-
that ( t, + t) A. Hence, (x , + t) A. But then + t S,
empty open as well as closed subsets of M. This contradicts the fact that
a contradiction , as sup S = . Hence, S is not bounded above. Thus,
(M, d) be a connected metric space. Hence, g(M ) is connected.
(x , ) A. Similarly, (, x + ) A. i.e R A R. Hence, A = R.
Proof 2. Let f : g(M ) {1} be a continuous function. Thus, f g :
Example 6.1 Note that R is not connected as (, 0) and (0, ) are M {1} be a continuous function and M is connected. Hence, f g is
open sets. Further, they are complements of each other. Hence, they are constant on M. Therefore f is constant on g(M ). Hence, g(M ) is connected.
closed sets. Thus, (, 0) is a non-empty proper subset of R which is
Corollary 6.2 [1, 1] is connected.
both open and closed. Hence, R is not connected.
Proof. Define f : R R by f (x) = sin x. Note that f is a continuous
Theorem 6.2 A metric space (M, d) is connected if and only if every con- function and R is connected. Hence, f (R) = [1, 1] is connected.
tinuous function f : M {1} is a constant function.
Example 6.2 Show that [0, ) and (0, ) are connected sets.
Proof. Let (M, d) be a connected metric space and f : M {1} be a
continuous function. We want to show that f is a constant function. If f is
Example 6.3 Define f : R R by f (x) = | sin x|. Is f a continuous
a non-constant function then it is onto. Let A = f 1 (1) and B = f 1 (1).
function? What is f (R)? What can you say about f (R)?
Since f is a continuous function and {1} as well as {1} are closed sets,
A = f 1 (1) and B = f 1 (1) are closed sets. Further A B = M and 1 One may also use sequential criteria to prove the continuity of f.
47
Example 6.4 Define f : R R by f (x) = 1+sin 2

x
. Is f a continuous 1. Let A and B be two connected subsets of M such that A B 6= .
function? What is f (R)? What can you say about f (R)? Then A B is connected.
Example 6.5 Define g : R R by g(x) = (a b)( 1+sin x

) + b. Is g a 2. Let A be a connected subset of the metric space M. If A B A.
2
continuous function? What is g(R)? What can you say about g(R)? Then B is connected.
3. Let {Ai }iI be a collection of the connected subsets of M such that
Theorem 6.4 The interval [a, b] is connected. S
for every i, j I, we have Ai Aj 6= . Then iI Ai is connected.
Proof 1. Using Example 6.5 we get the result.
Proof.
Proof 2. Assume that [a, b] is not connected.We then can write [a, b] =
U V, where U and V are non-empty proper open subsets of [a, b] such that 1. Let c A B and f : A B {1} be a continuous function. Since
U V = . Without loss of generality, we may assume that a U. A is connected, f is constant on A. Similarly B is connected, hence
Claim 1. V = . f is constant on B. Let b B. Hence, f (b) = f (c) for all b B. Also,
Since a U and U is open there exists an > such that [a, a + ) U. f (a) = f (c) for all a A. Thus, f (x) = f (c) for all x A B. Thus,
Hence, [a, a + 2 ] U. Let every continuous function f : A B {1} is a constant function.
Hence, A B is connected.
E = {x [a, b]|[a, x] U }. 2. Let f : B {1} be a continuous function. Since A B is a
connected subset of the metric space M, f is a constant on A, say 1.
Since, [a, a+ U, a+ E. Hence, E 6= . Further, E is bounded above
2] 2 Let b B. Since f is continuous at b, (given the open set {f (b)} in
by b. Hence, by LUB axiom, there exists c R such that c = sup E.. Note
{1}, there exists an open set U such that b U and f (U ) {f (b)}.
that a c b. We now claim that c E. For each n N, c n1 is not an
Since, b B A, b is a limit point of A. In particular, there exists
upper bound for E. Hence, there exists an xn E such that c n1 xn c.
a A U. So we find that f (a) {f (b)}, that is, f (a) = f (b) = 1.
Using Sandwich theorem, limn xn = c. Since, xn E, xn U. Since, U
Thus, f is constant on B. Hence B is connected.
is closed in [a, b] and c [a, b], we get that c U. Now,
S
3. Let f : A = iI Ai {1} be a continuous function and a A.
1
[a, c) = nN [a, c ] [a, xn ]. We want to prove that f (x) = f (a) for all x A.
n S
Since, a A and A = iI Ai , a Ai for some i I. Since, Ai is a
As each of [a, xn ] U, [a, c) U. Further, c U. Hence, [a, c] U and connected subset of M, f (x) = f (a) for all x Ai . Let j I, j 6= i.
hence c E. We now show that c = b and thus V = . Since, Ai Aj 6= , using (1), we get that Ai Aj is a connected subset
Since, c U and U is an open set, there exists an open set containing c of M. Hence, f (x) = f (a) for every x Ai Aj as f is constant on
lying in U. If c < b, there exists an integer N N such that (c N1 , c+ N1 ) Ai Aj for every j. Thus, f is constant on A. Hence A is connected.
1
U. Thus, c + 2N E, a contradiction, as sup E = c. Hence, c = b. Hence,
[a, b] is a connected set. Lemma 6.1 Let (M, d) be a metric space such that given any two points
x, y M there exists a connected set A such that x, y A. Then M is
Example 6.6 Define f : [0, 2] R2 by f (t) = (a + r cos t, b + r sin t), connected.
where r > 0. Note that f is a continuous function and [0, 2] is a connected Proof. Let f : M {1} be a continuous function. Suppose there exists
set. Hence, f ([0, 2]) is a connected set. But f ([0, 2]) = {(x, y)|(x a)2 + x, y M such that f (x) = 1 and f (y) = 1. By the given information, there
(y b)2 = r2 }. Thus, f ([0, 2]) is a circle with center (a, b) and radius r. exists a connected set A such that x, y A. Note that f |A is a continuous
Thus, a circle is a connected set. function and since A is a connected set, f |A is a constant function. Hence,
f (x) = f (y), a contradiction. Hence, f | is a constant function and M is a
Theorem 6.5 Let (M, d) be a metric space. connected set.
Theorem 6.6 (Connected subsets of R) A set J R is connected if assume that an = 1. If a0 = 0 then p(0) = 0 i.e 0 is a root of p(x). We may
and only if J is an interval. assume that a0 6= 0. Define
Proof. If J R is not an interval, then there exist x, y J and a z between
p(x) an1 a0
( that z
x and y such / J. Consider the function f : R {z} {1} defined q(x) = i.e. q(x) = 1 + + + n.
1 if s < z xn x x
by f (s) = . Then f is a non-constant continuous function.
1 if s > z. For x > 1, we have
Hence, f |J is a non-constant continuous function onto {1}. Hence, J R a
n1 a0 |an1 | |a0 |
is not connected. |q(x) 1| = + + n + + n
x x |x| |x |

To prove the converse, let J be any non-empty interval. Let x, y J.
n
Since J is an interval, [x, y] J. Further, [x, y] is a connected set. Thus, J |an1 | + + |a0 | A X
= , where A = |ai |
has the property that two of its points lie in a connected set. Hence, J is |x| |x| i=0
a connected set by the above lemma.
Thus, if |x| > max{1, 2A}, then |q(x) 1| < 12 . Hence, q(x) > 0 if |x| >
Theorem 6.7 (Intermediate Value Theorem) Let f : [a, b] R be a max{1, 2A}. If we now choose > max{1, 2A}, then q() and q() are
continuous function. Assume that y is a point between f (a) and f (b). Then both positive. But then p() is positive and p() is negative. Hence by
there exists x [a, b] such that f (x) = y. intermediate value theorem there exists [, ] such that p() = 0.
Proof. Note that [a, b] is connected. Since f is a continuous function on
[a, b], its image under f, that is, f [a, b] is connected. Since, any connected Example 6.7 Show that the annulus A = {(x, y)|1 < x2 + y 2 < 4} is
subset of R is an interval and f [a, b] is connected, f [a, b] is an interval, say connected.
J. Now, f (a), f (b) J. By the definition of an interval, the point y J, Solution 1. Note that [1, 2) and [0, 2] are connected subsets of R. Hence,
that is y f [a, b]. Therefore, there exists x [a, b] such that f (x) = y. [1, 2) [0, 2] are connected subsets of R2 . Define f : [1, 2) [0, 2] R2 by
f (r, ) = (r cos , r sin ). Note that f is a continuous function and image of
Theorem 6.8 (Existence of nth roots) Let [0, ) and n N. f is A. Since, [1, 2) [0, 2] are connected subsets of R2 , A is a connected
Then there exists a unique x [0, ) such that xn = . subset of R2 .
Proof. If = 0 then x = 0 and 0n = 0. So we assume that > 0. Solution 2. If we pick up any two points in A, then we can join them by
Consider the continuous function f : [0, ) R defined by f (t) = tn . a polygonal path which lies entirely in A. Hence, A is connected. 2
By Archimedean property of |R there exits a natural number N such that
N > . We have Theorem 6.10 Let X and Y be connected metric spaces. Then the prod-
uct space X Y is connected.
0 = f (0) < and f (N ) = N n N > . Proof. Let f : X Y {1} be a continuous function. Let (x0 , y0 )
X Y be fixed. Let (x, y) X Y be an arbitrary point in X Y. If we
Hence, using Intermediate value theorem to the interval [0, N ], we see that
show that f (x, y) = f (x0 , y0 ) then we are done.
there exists x [0, N ] such that xn = .
To prove the above claim, let us first observe that iy : X X Y
Suppose there exists y > 0 such that y n = . We may assume that
defined by iy (x) = (x, y) is a continuous function. Similarly, the map
x < y. Hence, xn < y n < , a contradiction. Hence, x is unique.
ix : Y X Y defined by ix (y) = (x, y) is a continuous function. Therefore
for every point y Y, the subset X {y} = {(x, y)|x X} is a connected
Theorem 6.9 Any polynomial with real coefficients and of odd degree has
subset of X Y ; similarly, the subset {x} Y = {(x, y)|x X} is a
a real root.
connected subset of X Y for every point x X.
Proof. Let p(x) = an xn + an1 xn1 + + a1 x + a0 , where ai R and
an 6= 0 and n is odd. Since the roots of p(x) and a1n p(x) are same, we may 2 This solution is better than the first, as we can easily generalize this.
Now the point (x, y0 ) lies in both X {y0 } and {x} Y. The restriction
of f to either of these sets are continuous and hence constants as the subsets
X {y} and X {y0 } are connected. Hence, f (x, y0 ) = f (x, y) for every
y Y and f (x, y0 ) = f (x0 , y0 ) for every x X. Thus, f (x, y) = f (x, y0 ) =
f (x0 , y0 ) for every x X and y Y. Thus, f is constant. Hence, X Y is
connected.
Example 6.8 Show that a circle or a line or a parabola is not homeomor-

phic in R2 is not homeomorphic to a hyperbola.
Solution. Note that a circle or a line or a parabola is a connected subset
of R2 but a hyperbola is not a connected subset of R2 . Under a continuous
function, a continuous function image of a connected subset is a connected
subset. Hence, the result.
Example 6.9 Let X be the union of axes given by xy = 0 in R2 . Is it

homeomorphic to a line, a circle, a parabola or a rectangular hyperbola
xy = 1?
Solution. X is not compact while a circle is compact. Hence, X is not
homeomorphic to a circle. X is connected but a rectangular hyperbola
xy = 1 is not connected. Hence, X is not homeomorphic a rectangular
hyperbola xy = 1.
Suppose there exists a f : X S, where S is either a line, a circle
or a parabola. If you remove (0, 0) from X then we get 4 connected parts
however if we remove f (0, 0) from S i.e. a point from S then we get at most
two connected parts. Hence there is no homeomorphism from X to S.
Index
Algebra of limits, 25
Appendix AM-GM Inequality, 6
We say that any two norms are equivalent if each one is less than or equal
to a positive constant times the other. Cauchy sequence, 21
Let V be a finite dimensional vector space over R. Then all norms on Cauchy-Schwarz Inequality, 2
V are equivalent. closed ball, 15
Proof. Let {v1 , v2 , . . . , vn } be a basis of V and let k k be the sup norm cluster point, 19
with respect to this basis. We let | | be any other norm. If x V and Convergent sequence, 20
xi are its coordinates with respect to this basis i.e. x = x1 v1 + + xn vn .
Hence, Holders Inequality, 8
|v| = |x1 v1 + + xn vn | C kxk ,
interior point, 16
where C = n sup ki k. This proves one inequality, and shows that the norm
| | is continuous with respect to k k . Hence, | | has a minimum on the limit of a function, 25
unit sphere with respect to k k , say at the point v V, so that limit point, 20
|v| |x|, for all x V, kxk = 1. metric space, 11

Minkowskis Inequality, 9
Let y V, y 6= 0 and write
open ball, 15
y = y1 1 + + yn n , yi R.
Youngs Inequality, 7
Let j be such that |yj | = max |yi | = kyk . Then y = yi x with kxk = 1, and
|v| |x| = |y/yj | = |y|/|yj |.
It follows that
|v| kyk |y|,
thus proving the other inequality, and completing the proof of our assertion.
55 56

Metric Spaces

Uploaded by

Copyright:

Available Formats

Metric Spaces

Uploaded by

Document Information

Original Description:

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Metric Spaces

Uploaded by

Copyright:

Available Formats

ii Dr. V. V.

Topology of Metric Spaces

January 14, 2015

4 Limits in Metric spaces 25 Dr. V. V. Acharya

5 Compact Metric Spaces 33

2. kkxk = |k| kxk

R, Rn and C, Cn Exercise 1.1 Let a1 , . . . , an ; b1 , . . . , bn be real numbers. Find the discrim-

Thus the distance d(A, B) in C is the same as between the corresponding

ma1 < a1 + a2 + . . . + am < mam , f (x0 ) = + x0 y = + y p1 y (1.15)

7. Let M = C. Define d(z, w) = |z w|. It is easy to see that (C, d)

4. As in (3) but with d(x, y) = | log(x/y)|.

10. Let the mapping d : R2 R2 R be defined by d(x, y) = max{|x1 Solutions

15. Let X be the set of all continuous

Exercise 3.2 Determine B((0, 0); 1) in the following metric spaces:

3.1 Open Balls Exercise 3.4

open subset of M if every point of G is an interior point.

d(a, y) d(a, x) + d(x, y) s + t < r. S = {I G|I is an open interval containing x}.

4.1 Continuous functions

0 < d1 (x, a) < and x E, d2 (f (x), L) < . suffices to show that

of A which is -dense in A. tradicts our hypothesis. Hence, A must be totally bounded.

Corollary 5.2 Let f be a continuous function from the compact metric

(b) {(x, y, z)|x2 y 2 + z 2 = 1}

A B = . Thus, A and B are open sets. Thus, A and B are non-empty

Example 6.4 Define f : R R by f (x) = 1+sin 2

Example 6.5 Define g : R R by g(x) = (a b)( 1+sin x

Example 6.8 Show that a circle or a line or a parabola is not homeomor-

Example 6.9 Let X be the union of axes given by xy = 0 in R2 . Is it

|v| |x|, for all x V, kxk = 1. metric space, 11

|v| |x| = |y/yj | = |y|/|yj |.

You might also like